As a long-term goal, robots should be able to interact with humans in multiple forms in order to assist them in everyday life. Over the course of several projects with public operational environments, for our mobile robots equipped with touch displays as primary input/output device, we found an intuitive way to interact with the robot is in a seated position and being approached by the robot, so that the user can physically operate the robot. Sitting down is also a signal for interaction intention which is easily conveyed and even visible from afar, as opposed to speech recognition systems. To realize this behavior, we propose a succinct yet effective method of recognizing a seated person by utilizing the height estimation of a person detector, as well as a method of finding an interaction pose to approach the user and navigating to that pose even in dynamic unstructured environments. The proposed approaches are evaluated in an experimental setup similar to the dynamic hospital environment of our project ROGER where a robotic gait training coach for orthopedic rehabilitation of patient with hip prosthesis is developed.
展开▼
